Pith. sign in

REVIEW 4 major objections 6 minor 42 references

Adaptive Target Localization under Uncertainty using Multi-Agent Deep Reinforcement Learning with Knowledge Transfer

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A single multi-agent model handles reachable, unreachable, and absent targets in one shared network.

desk verdict A credible new task formulation for multi-agent localization under existence/reachability uncertainty, but the benchmark comparison is structurally biased and the reward function is undefined for one of the three scenarios. read the letter →

arxiv 2501.10924 v1 pith:W4COQC2K submitted 2025-01-19 cs.LG cs.AIcs.RO

Ahmed Alagha , Rabeb Mizouni , Shakti Singh , Jamal Bentahar , Hadi Otrok This is my paper · ORCID
classification cs.LGcs.AIcs.RO
keywords Multi-AgentDeepReinforcementLearningTargetlocalizationunderuncertaintyProximalPolicyOptimizationTransferSwarmroboticsFalsealarmsunreachabilityRadiation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that one multi-agent deep-reinforcement-learning framework can handle the three situations a target-localization swarm actually meets: the target is reachable, the target exists but cannot be reached, and no target exists at all (a false alarm). The authors argue that their PPO-trained CNN policies, fed with compact 2D observation maps, learn to move, declare existence, and declare reachability as one decision process, and that a transfer-learning step lets the same feature extractor estimate the coordinates of an unreachable target. The payoff claimed is faster and cheaper searches than a uniform sweep, a centralized DDQN, and an earlier MADRL baseline, at space complexity that stays fixed per agent.

What carries the argument

The load-bearing mechanism is an actor-critic CNN trained with PPO under centralized learning and distributed execution: each agent runs its own copy of the actor on ten reduced observation maps (nine local and global 2D maps plus a convolutional-autoencoder embedding of the environment layout), while a centralized critic stabilizes training. A team-based shaped reward, Eq. (5), gives all agents the same feedback based on whether the BFS shortest-path distance from the team to the true target decreased and how many agents moved. Once the policy learns to flag a target as unreachable, transfer learning copies the frozen convolutional layers into a regression head that outputs $(x,y)$ coordinates, so the deployed model is one network with two output heads.

What would settle it

Retrain with Eq. (5) replaced by a reward computable from observations alone, such as one based only on the readings map and the agents' positions, then compare episode length and cost on the same three scenario types; if performance collapses, the oracle distance is doing the work. Separately, instrument the no-target scenario by running episodes with no source and recording how often the team correctly declares absence within the step limit—the paper reports no such measure, and Eq. (5) does not specify a reward for it.

Watch

Extended reading notes

Core claim

The central claim is that the combined model—a shared convolutional feature extractor with one policy head for movement, detection, and reachability and one regression head for target coordinates—solves target localization under uncertainty better than the benchmarks. In Section 5.4 the paper states that the proposed work outperforms all benchmarks by achieving faster and less costly localization; Section 5.3 reports that the training and validation loss of the estimation model converge to zero, which the paper reads as accurate location estimates. The authors attribute the gains to PPO with a centralized critic, a shaped team reward that uses BFS shortest-path distances to reward progress, and transfer learning that avoids training the estimator from scratch.

Load-bearing premise

The load-bearing premise is that the team reward in Eq. (5)—which feeds the true target's BFS shortest-path distance to the agents during training—is a legitimate training signal, and that the no-target scenario, whose reward the paper never defines, is handled by the same rule; if either fails, the claimed convergence and benchmark wins do not transfer to deployment.

Editorial extensions

If this is right

  • A 4-agent team using the trained policy can finish reachable, unreachable, and absent-target episodes within the 100-step limit, with the strongest target taking about 7.7 steps and 12.4 movement actions on average.
  • Because the actor's state and action spaces are fixed per agent, adding more agents does not increase the space complexity beyond $O(I)$; the same policy copy is deployed on each agent.
  • Since the convolutional feature extractor is shared with the estimation head, triggering an unreachable-target estimate does not require a second network or extra per-agent parameters.
  • Against the reported benchmarks, the method is claimed to dominate Uniform search, centralized DDQN, and ODMTL on both episode length and movement cost under all tested target strengths.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's own results suggest a harder test the authors did not run: training with an oracle-free reward and measuring false-alarm performance, since the claimed behavior in the absent-target scenario has no stated reward signal.
  • If the framework transfers as claimed to other inverse-square-law signals such as sound or heat, the same two-head architecture and observation maps should work with only the forward sensor model changed; this is a natural benchmark the paper does not provide.
  • The validation loss converging to zero for coordinate estimation may indicate the estimation subtask is nearly saturated by the privileged BFS-based observations; a more informative evaluation would report meter-level error distributions, not just loss curves.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes a multi-agent deep reinforcement learning method for target localization in uncertain environments, where the target may be reachable, unreachable, or absent. The method uses PPO with CNN-based actor-critic networks under centralized learning and decentralized execution, a team-shaped reward, and an action space spanning mobility, existence detection, and reachability declaration. When the target is declared unreachable, a separate deep learning model, trained via transfer learning from the MADRL actor, estimates the target location using the shared feature extractor. The approach is evaluated in a simulated radiation localization environment with varying team sizes and target strengths, and is compared against Uniform search, DDQN, and the authors' prior ODMTL method. The paper also provides a complexity analysis showing O(I) space complexity per agent for the proposed method.

Significance. If validated, the work would be a useful extension of MADRL-based target localization to practically important scenarios that previous methods assume away: false alarms and unreachable targets. The action-space decomposition into mobility, existence, and reachability is a natural formalization, and the use of transfer learning to share a feature extractor between the decision policy and the estimation model is a sensible design for reducing computational overhead. The complexity analysis in Section 5.5 correctly highlights the scalability advantage of distributed action/state spaces over centralized DDQN. However, the empirical evidence currently has several gaps that directly affect the central claims: the reward function is underspecified for two of the three scenarios, the benchmark comparison is structurally biased, and no statistical variability is reported. These issues are fixable, but they are load-bearing for the claimed outperformance.

major comments (4)
  1. [Section 4.4, Eq. (5)] The reward function is not well-defined in two of the three scenarios that the paper explicitly targets. In the no-target scenario there is no target, so the set D_t and hence min(D_t) cannot be computed; in the unreachable-target scenario, if 'unreachable' means that no path exists, the BFS shortest-path distance is also undefined. Since Section 5.1 states that each episode is randomly assigned to one of the three scenarios with equal probability, the reward signal for two-thirds of training episodes is unspecified. The paper must state the reward for correct and incorrect existence/reachability flags and for episodes with no finite path to the target.
  2. [Section 5.4, Fig. 9] The benchmark comparison is not apples-to-apples. Uniform search, DDQN, and ODMTL have no actions to declare the target non-existent or unreachable, so in the no-target and unreachable-target episodes (two-thirds of the test distribution per Section 5.1) they are forced to run until the 100-step limit, while the proposed method can terminate early through its majority flags. Thus a large part of the time and cost gap in Fig. 9 is incurred by construction rather than by learned superiority. The general claim that the proposed work 'outperforms all the benchmarks' should be supported by scenario-disaggregated results (e.g., reachable-target episodes only) or narrowed to 'the only method that can handle all three scenarios.'
  3. [Section 5.2, Section 5.4, Figs. 6-9] No error bars, confidence intervals, or multiple-seed results are reported anywhere in the evaluation. The learning curves and benchmark comparisons appear to come from single training runs or single evaluation averages, so the magnitude of the reported improvements over baselines and the differences across team sizes cannot be assessed statistically. Please report means and standard deviations over at least five independent seeds, and where appropriate include significance tests.
  4. [Section 5.3, Fig. 8] The claim that the target location is 'estimated accurately' rests entirely on training and validation loss curves that 'converge to 0.' No localization error in physical units, no accuracy threshold, and no evaluation across different environment layouts is reported. Since the estimation head is a central component of the combined model, the paper should report a quantitative error metric, such as mean Euclidean distance between the estimated and true target coordinates on held-out episodes.
minor comments (6)
  1. [Abstract] The phrase 'a team-based reward functions is proposed' is ungrammatical and should read 'a team-based reward function is proposed.'
  2. [Section 4.2] The text says the five observations are reduced into 'ten optimized ones,' but the actor is described as taking 'the first 9 reduced observations' plus a CAE embedding; please clarify whether the CAE embedding is the tenth observation and how the nine are selected.
  3. [Table 2] The state-space dimensionality for the proposed method is listed as 9I, but the actor also receives the CAE embedding of dimension d=128; the table should include this term or justify its omission.
  4. [Section 5.5] The FLOPs and trainable-parameter counts are reported without specifying whether they include the estimation head and the CAE encoder; please state the exact configuration used for these numbers.
  5. [Eq. (4)] If the direction index d_i ranges over {1,...,D}, then d_i=1 corresponds to angle 2π/D rather than 0; please clarify whether directions are indexed from 0 or the angle formula is adjusted accordingly.
  6. [Section 5.1] No information is given about random seeds, environment-generator reproducibility, or code release; providing these would substantially strengthen reproducibility of the reported curves.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the RL reward/evaluation loop and the transfer-learning estimator are empirical pipelines; the only self-citations are baseline descriptions, not load-bearing premises.

full rationale

The paper contains no derivation that reduces to its own inputs. The MADRL policy is optimized with PPO using the shaped team reward in Eq. (5), where min(D_t) is computed by BFS from the true target; this is privileged training feedback, but it is not a fitted constant later relabeled as a prediction, and the reported evaluation metrics (episode length and movement cost) are measured independently in the environment. The target-estimation head in Section 4.5 is trained as a supervised regressor on observations paired with true coordinates, with the MADRL features transferred and frozen; reporting training/validation MSE convergence is a standard held-out evaluation rather than a circular claim. The self-citations to [4,9] describe the ODMTL baseline and prior MADRL target-localization work; they are not invoked as uniqueness theorems or as evidence for the central claim. The benchmark comparison in Section 5.4 is arguably unfair because Uniform, DDQN, and ODMTL have no actions to declare target non-existence or unreachability, so in the one-third no-target and one-third unreachable episodes they are forced toward the 100-step limit; that is a comparative-validity concern, not circularity. Overall, the strengths of the claims rest on empirical measurements, not on equations that are equivalent to their own inputs.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a privileged training-time reward oracle (Eq. 5), a monotonic inverse-square sensor model (Eq. 6), lossless agent communication, and an empirical transfer-learning assumption. Five hand-chosen hyperparameters shape the reported results. No new physical entities are postulated.

free parameters (5)
  • Q (wrong-flag penalty) = 500
    Penalty applied when agents incorrectly flag target non-existence or unreachability (Eq. 5, Section 4.4); chosen by hand with no sensitivity analysis.
  • D (number of movement directions) = 8
    Section 4.3: 'discretizing the direction of movement into 8 possible values is sufficient'; chosen by hand.
  • n (reduced observation window size) = 7
    Section 4.2 declares n a tunable hyperparameter; Section 5.5 reports 7x7 maps with I = 441. No ablation on n is provided.
  • CAE embedding size = 128
    Table 1 sets the environment-layout embedding to 128; no ablation on the embedding dimension is reported.
  • Obstacle attenuation factor mu = 0.1
    Section 5.1: 'constant attenuation of mu = 0.1 for each obstacle'; an environment constant chosen for the simulation that shapes all sensor readings.
assumptions (6)
  • domain assumption Radiation readings follow the inverse-square law CPM proportional to S/d^2 with constant per-obstacle attenuation (Eq. 6, Section 5.1).
    The observation design and the claim of generality to 'any localization environment' rest on readings being a monotonic function of distance to the target.
  • domain assumption Agents can communicate and share information, and store previously collected observations (Section 4.2).
    Team distribution, visit history, and readings history maps assume lossless sharing and storage; no communication model or cost is simulated.
  • standard math PPO with a centralized critic (CLDE) handles the multi-agent non-stationarity (Section 4.4).
    Background result cited to [17,37]; accepted as standard in the MARL literature.
  • domain assumption Feature representations learned by the MADRL actor transfer to the target estimation task, so freezing the early layers and training only the last layer is sufficient (Section 4.5).
    The transfer-learning claim is validated only by the reported loss curves; no comparison against training the estimator from scratch is given.
  • domain assumption The CAE encoder generates embeddings that preserve the environment layout information needed for planning (Section 4.2).
    The CAE is trained on a synthetic layout dataset; the paper does not quantify reconstruction error on the test layouts used during MADRL evaluation.
  • ad hoc to paper The training-time reward oracle (Eq. 5) is well-defined in all three scenarios and is an admissible training signal.
    min(D_t) requires the true target location; in the no-target scenario D_t is undefined, and the paper never specifies the reward for false-alarm episodes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Target Localization under Uncertainty using Multi-Agent Deep Reinforcement Learning with Knowledge Transfer." pith.science (2026). https://pith.science/paper/W4COQC2K

@misc{pith2026250110924,
  author       = {Pith},
  title        = {Pith review of: Adaptive Target Localization under Uncertainty using Multi-Agent Deep Reinforcement Learning with Knowledge Transfer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W4COQC2K}},
  note         = {Machine review of arXiv:2501.10924}
}
read the original abstract

Target localization is a critical task in sensitive applications, where multiple sensing agents communicate and collaborate to identify the target location based on sensor readings. Existing approaches investigated the use of Multi-Agent Deep Reinforcement Learning (MADRL) to tackle target localization. Nevertheless, these methods do not consider practical uncertainties, like false alarms when the target does not exist or when it is unreachable due to environmental complexities. To address these drawbacks, this work proposes a novel MADRL-based method for target localization in uncertain environments. The proposed MADRL method employs Proximal Policy Optimization to optimize the decision-making of sensing agents, which is represented in the form of an actor-critic structure using Convolutional Neural Networks. The observations of the agents are designed in an optimized manner to capture essential information in the environment, and a team-based reward functions is proposed to produce cooperative agents. The MADRL method covers three action dimensionalities that control the agents' mobility to search the area for the target, detect its existence, and determine its reachability. Using the concept of Transfer Learning, a Deep Learning model builds on the knowledge from the MADRL model to accurately estimating the target location if it is unreachable, resulting in shared representations between the models for faster learning and lower computational complexity. Collectively, the final combined model is capable of searching for the target, determining its existence and reachability, and estimating its location accurately. The proposed method is tested using a radioactive target localization environment and benchmarked against existing methods, showing its efficacy.

Figures

Figures reproduced from arXiv: 2501.10924 by the authors.

Figure 1
Figure 1. Three examples showing the different scenarios to be addressed by the agents, including cases of (a) complex environments with obstacles, (b) unreachable targets, and (c) no targets due to false alarms. In all scenarios, it is desired for the task to be executed as quickly as possible and with low resource consumption, which requires the agents to cooperate. Although the target location is unknown, the agents can us… view at source ↗
Figure 2
Figure 2. An overview of the model proposed, which is to be deployed on each sensing agent. 4.1. MADRL Formulation and Policy Optimization In the context of MADRL, Markov Games are generally used to extend Markov Decision Processes (MDPs) into multi-agent settings [30, 31, 32]. In the problem of target localization, the state at a given instant is defined by the distribution of agents and obstacles, as well as the target loca… view at source ↗
Figure 3
Figure 3. The five collected observations by a sensing agent in a team of three agents. The original observations (top row) are processed to obtain the reduced observations (bottom row). The reduced observations are either local (green) or global (orange). It is worth mentioning that due to temporal aspect of the problem, using methods in Recurrent Neural Networks (RNNS), such as Gated Recurrent Units (GRUs) and Long Short-Te… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: The actor architecture trained through MADRL (top), and the architecture of the estimation model trained through DL and TL (bottom). Execution (CLDE) [37]. Here, a copy of the actor is given to each agent, where agents act independently based on their observations. Dur…
Figure 5
Figure 5. Figure 5: The final model deployed on each of the sensing agents. 5. Simulation and Evaluation Extensive experiments are conducted in this section to validate the efficiency of the proposed method, as well as benchmark it against existing works in the literature. All the simulat…
Figure 6
Figure 6. Figure 6: A summary of the training plots for different team sizes and for varying target strengths. The episodic reward is shown in (a)-(c), the episodic length is shown in (d)-(f), and the episodic cost is shown in (g)-(i). area. On the other hand, the similar performance acro…
Figure 7
Figure 7. Figure 7: The agents’ performance under the different scenarios in terms of (a) episodic time and (b) episodic cost, for a team of 4 agents and varying target strengths. for target estimation. Here, a dataset is built combining the observations collected by the MADRL agents unti…
Figure 8
Figure 8. Figure 8: The training and validation loss for the target estimation model using TL. 5.4. Benchmarks This section compares the performance of the proposed MADRL method with some existing benchmarks in the literature. As discussed in Section 1, existing meth￾ods struggle when add…
Figure 9
Figure 9. Figure 9: Comparison between the performance of the proposed method and the benchmarks in terms of (a) episode length and (b) episode cost. ODMTL [9] [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 32 canonical work pages

  1. [1]

    M. C. De Donato, F. Corradini, F. Fornari, B. Re, Safe: An ict platform for supporting monitoring, localization and rescue operations in case of earthquake, Internet of Things 27 (2024) 101273

    work page 2024

  2. [2]

    L. A. Hussain, S. Singh, R. Mizouni, H. Otrok, E. Damiani, A predictive target tracking framework for iot using cnn–lstm, Internet of Things 22 (2023) 100744

    work page 2023

  3. [3]

    Alagha, S

    A. Alagha, S. Singh, R. Mizouni, A. Ouali, H. Otrok, Data-driven dynamic active node selection for event localization in IoT applications-a case study of radiation localization, IEEE Access 7 (2019) 16168–16183

    2019

  4. [4]

    Alagha, S

    A. Alagha, S. Singh, R. Mizouni, J. Bentahar, H. Otrok, Target localization us- ing multi-agent deep reinforcement learning with proximal policy optimization, Future Generation Computer Systems 136 (2022) 342–357. 22

    work page 2022

  5. [5]

    M. N. Ramadan, T. Basmaji, A. Gad, H. Hamdan, B. T. Akg ¨un, M. A. Ali, M. Alkhedher, M. Ghazal, Towards early forest fire detection and prevention using ai-powered drones and the iot, Internet of Things (2024) 101248

    work page 2024

  6. [6]

    H. B. Salameh, M. Alhafnawi, A. Masadeh, Y . Jararweh, Federated reinforcement learning approach for detecting uncertain deceptive target using autonomous dual uav system, Information Processing & Management 60 (2023) 103149

    work page 2023

  7. [7]

    Charef, A

    N. Charef, A. B. Mnaouer, M. Aloqaily, O. Bouachir, M. Guizani, Artificial intelligence implication on energy sustainability in internet of things: A survey, Information Processing & Management 60 (2023) 103212

    work page 2023

  8. [8]

    Grassi, M

    L. Grassi, M. Ciranni, P. Baglietto, C. T. Recchiuto, M. Maresca, A. Sgorbissa, Emergency management through information crowdsourcing, Information Pro- cessing & Management 60 (2023) 103386

    work page 2023

Show all 42 references
  1. [9]

    Alagha, R

    A. Alagha, R. Mizouni, J. Bentahar, H. Otrok, S. Singh, Multi-agent deep re- inforcement learning with demonstration cloning for target localization, IEEE Internet of Things Journal (2023)

    2023

  2. [10]

    Z. Liu, S. Abbaszadeh, Double Q-learning for radiation source detection, Sensors 19 (2019) 960

    2019

  3. [11]

    E.-w. Bai, K. Yosief, S. Dasgupta, R. Mudumbai, The maximum likelihood es- timate for radiation source localization: Initializing an iterative search, in: 2014 53rd IEEE Conference on Decision and Control, IEEE, pp. 277–282

    2014

  4. [12]

    J.-C. Chin, D. K. Yau, N. S. Rao, Y . Yang, C. Y . Ma, M. Shankar, Accurate local- ization of low-level radioactive source under noise and measurement errors, in: Proceedings of the 6th ACM conference on Embedded network sensor systems, ACM, pp. 183–196

  5. [13]

    Proctor, C

    P. Proctor, C. Teuscher, A. Hecht, M. Osi ´nski, Proximal policy optimization for radiation source search, Journal of Nuclear Engineering 2 (2021) 368–397

    2021

  6. [14]

    Shurrab, R

    M. Shurrab, R. Mizouni, S. Singh, H. Otrok, Reinforcement learning frame- work for uav-based target localization applications, Internet of Things 23 (2023) 100867

    2023

  7. [15]

    R. S. Sutton, A. G. Barto, Reinforcement learning: An introduction, MIT press, 2018

    2018

  8. [16]

    T. T. Nguyen, et al., Deep reinforcement learning for multiagent systems: A re- view of challenges, solutions, and applications, IEEE transactions on cybernetics 50 (2020) 3826–3839

    2020

  9. [17]

    Schulman, et al., Proximal policy optimization algorithms, arXiv preprint arXiv:1707.06347 (2017)

    J. Schulman, et al., Proximal policy optimization algorithms, arXiv preprint arXiv:1707.06347 (2017). 23

    2017 arXiv

  10. [18]

    Ziock, W

    K. Ziock, W. Goldstein, The lost source, varying backgrounds and why bigger may not be better, in: AIP Conference Proceedings, volume 632, American Insti- tute of Physics, pp. 60–70

  11. [19]

    A. H. Liu, J. J. Bunn, K. M. Chandy, An analysis of data fusion for radiation detection and localization, in: 2010 13th International Conference on Information Fusion, IEEE, pp. 1–8

    2010

  12. [20]

    H. Xiao, R. Cui, D. Xu, A sampling-based bayesian approach for cooperative multiagent online search with resource constraints, IEEE Transactions on Cyber- netics 48 (2017) 1773–1785

    2017

  13. [21]

    Sartoretti, J

    G. Sartoretti, J. Kerr, Y . Shi, G. Wagner, T. S. Kumar, S. Koenig, H. Choset, Primal: Pathfinding via reinforcement and imitation multi-agent learning, IEEE Robotics and Automation Letters 4 (2019) 2378–2385

    2019

  14. [22]

    S. K. Pandey, M. A. Zaveri, Event localization in the internet of things environ- ment, Procedia computer science 115 (2017)

    2017

  15. [23]

    Grondin, et al., Sound event localization and detection using crnn on pairs of microphones, in: 2019 Proc

    F. Grondin, et al., Sound event localization and detection using crnn on pairs of microphones, in: 2019 Proc. Detection and Classification of Acoustic Scenes Events Workshop

    2019

  16. [24]

    Z. Liu, W. Dai, M. Z. Win, Node placement for localization networks, in: 2017 IEEE International Conference on Communications (ICC), IEEE, pp. 1–6

    2017

  17. [25]

    Alagha, S

    A. Alagha, S. Singh, H. Otrok, R. Mizouni, RFLS-resilient fault-proof localiza- tion system in IoT and crowd-based sensing applications, Journal of Network and Computer Applications 170 (2020)

    2020

  18. [26]

    Alagha, R

    A. Alagha, R. Mizouni, S. Singh, H. Otrok, A. Ouali, SDRS: A stable data- based recruitment system in IoT crowdsensing for localization tasks, Journal of Network and Computer Applications 177 (2021) 102968

    2021

  19. [27]

    Alagha, S

    A. Alagha, S. Singh, H. Otrok, R. Mizouni, Influence-and interest-based worker recruitment in crowdsourcing using online social networks, IEEE Transactions on Network and Service Management (2022)

    2022

  20. [28]

    Shurrab, S

    M. Shurrab, S. Singh, R. Mizouni, H. Otrok, IoT sensor selection for target localization: A reinforcement learning based approach, Ad Hoc Networks 134 (2022) 102927

    2022

  21. [29]

    Lazna, P

    T. Lazna, P. Gabrlik, T. Jilek, L. Zalud, Cooperation between an unmanned aerial vehicle and an unmanned ground vehicle in highly accurate localization of gamma radiation hotspots, Int. Journal of Advanced Robotic Systems 15 (2018) 1729881417750787

    2018

  22. [30]

    Gronauer, K

    S. Gronauer, K. Diepold, Multi-agent deep reinforcement learning: a survey, Artificial Intelligence Review (2021) 1–49. 24

    2021

  23. [31]

    Alagha, J

    A. Alagha, J. Bentahar, H. Otrok, S. Singh, R. Mizouni, Blockchain-assisted demonstration cloning for multi-agent deep reinforcement learning, IEEE Inter- net of Things Journal (2023)

    2023

  24. [32]

    Alagha, H

    A. Alagha, H. Otrok, S. Singh, R. Mizouni, J. Bentahar, Blockchain-based crowd- sourced deep reinforcement learning as a service, Information Sciences (2024) 121107

    2024

  25. [33]

    Schulman, P

    J. Schulman, P. Moritz, S. Levine, M. Jordan, P. Abbeel, High-dimensional con- tinuous control using generalized advantage estimation, in: 2016 Proc. Int. Conf. on Learning Representations (ICLR)

    2016

  26. [34]

    J. Sola, J. Sevilla, Importance of input data normalization for the application of neural networks to complex industrial problems, IEEE Transactions on nuclear science 44 (1997) 1464–1468

    1997

  27. [35]

    Damani, Z

    M. Damani, Z. Luo, E. Wenzel, G. Sartoretti, Primal 2: Pathfinding via reinforce- ment and imitation multi-agent learning-lifelong, IEEE Robotics and Automation Letters 6 (2021) 2666–2673

    2021

  28. [36]

    LeCun, et al., Lenet-5, convolutional neural networks, URL: http://yann

    Y . LeCun, et al., Lenet-5, convolutional neural networks, URL: http://yann. lecun. com/exdb/lenet 20 (2015) 14

    2015

  29. [37]

    Baker, I

    B. Baker, I. Kanitscheider, T. Markov, Y . Wu, G. Powell, B. McGrew, I. Mordatch, Emergent tool use from multi-agent autocurricula, in: 2020 Proc. Int. Conf. on Learning Representations (ICLR)

    2020

  30. [38]

    X. Lyu, Y . Xiao, B. Daley, C. Amato, Contrasting centralized and decentralized critics in multi-agent reinforcement learning, in: Proc. of the 2021 20th Interna- tional Conference on Autonomous Agents and MultiAgent Systems (AAMAS), pp. 844–852

    2021

  31. [39]

    Cheng, D

    Z. Cheng, D. Ye, T. Zhu, W. Zhou, P. S. Yu, C. Zhu, Multi-agent reinforcement learning via knowledge transfer with di fferentially private noise, International Journal of Intelligent Systems 37 (2022) 799–828

    2022

  32. [40]

    S. Shen, D. Ye, T. Zhu, W. Zhou, Privacy preservation in deep reinforcement learning: A training perspective, Knowledge-Based Systems 304 (2024) 112558

    2024

  33. [41]

    G. F. Knoll, Radiation detection and measurement, John Wiley & Sons, 2010

    2010

  34. [42]

    Davis, E

    D. Davis, E. Patronis, Sound system engineering, CRC Press, 2014. 25

    2014

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.